A study on drastic improvement of optical fiber transmission performance using deep learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 20K04532
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 21040:Control and system engineering-related
• Research Institution: Tokyo Institute of Technology
• Principal Investigator: Uenohara Hiroyuki 東京工業大学, 科学技術創成研究院, 教授 (20334526)
• Project Period (FY): 2020-04-01 – 2023-03-31
• Keywords: 光ファイバ通信 / 非線形歪補償 / 畳み込みニューラルネットワーク

Outline of Final Research Achievements

We aimed at improving the nonlinear compensation performance drastically for optical fiber transmission by using high recognition feature of deep learning compared with that of the conventional digital backpropagation method. First, for the reference, ANN was used, and the effective conditions of the sampling rate and the simultaneous number of input symbols, the number of neurons, and the number of epochs were obtained. Next, two-dimensional CNN was investigated, and input data were assigned to In-phase and Quadrature-phase components with time transient information. Introduction of majority vote was found to improve the performance. We have found that the performance of two-dimensional CNN indicating on complex plane or using I-Q-components in parallel both are superior to that of linear compensation only, in the same level compared with DBP with 1step/span, but in the middle between DBP with 1step/span and 2steps/span. Layer construction should be investigated further.

Free Research Field

光信号処理

Academic Significance and Societal Importance of the Research Achievements

光ファイバ通信システムの受信光信号の線形・非線形歪の補償性能を既存デジタル信号処理手法よりも抜本的に向上可能な、深層学習導入によるリアルタイム処理手法の実現を目指している。対象抽出能力の高い深層学習を導入し、既存のデジタル逆伝搬法の性能を改善する可能性を追究した。対象シンボル前後の情報による影響を事前学習させ、サンプリングレートと連続入力シンボル数を考慮に入れて、サンプリング後に信号再構築過程を導入し、人工ニューラルネットワーク、１次元あるいは２次元畳み込みニューラルネットワークにおいて、限定的ではあるが既存のデジタル逆伝搬法よりも補償性能の高い可能性を見出している。